Intelligent performance prediction and early warning system and method for roller compacted concrete adapted to strong wind environment

Abstract

The application discloses a kind of intelligent prediction and early warning system and method of roller compacted concrete performance suitable for strong wind environment.The system includes three layers of framework: edge intelligent collaborative monitoring layer dynamically collects multi-physical field and working condition data as needed;Strong wind special algorithm processing layer constructs five-dimensional time series dataset, accurately predicts the evolution of physical field through coupling type space-time prediction model, quantifies risk root cause and contribution ratio combined with Bayesian network causal diagram, accurately predicts the evolution of physical field and evaluates the root cause of later performance;Causal dynamic early warning application layer triggers multi-level early warning according to risk assessment results, generates early warning report containing risk root cause and pushes adaptive control scheme.The method includes five steps of construction preparation, dynamic acquisition preprocessing, coupling prediction analysis, dynamic early warning control and adaptive iterative optimization.The application improves the prediction accuracy and acquisition reliability under extreme wind conditions, realizes early risk prediction and root cause positioning, significantly improves the project quality compliance rate and reduces the control cost.

Description

Technical Field

[0001] This invention relates to the field of quality control and intelligent early warning of roller-compacted concrete construction in water conservancy and hydropower projects, specifically to an intelligent prediction and early warning system and method for roller-compacted concrete performance adapted to strong wind environments. Background Technology

[0002] In major projects such as water conservancy and hydropower, and high-altitude airports, large-volume concrete structures like roller-compacted concrete (RCC) often face complex construction environments with strong winds and low temperatures. Wind not only accelerates surface moisture evaporation and causes uneven internal humidity distribution in RCC, but also alters the hydration heat release rate through convective heat transfer, thus affecting the strength development, crack resistance, and long-term durability of the RCC. Furthermore, compared to ordinary concrete, RCC typically employs a dry-hard mix design with large-graded aggregates and low cement content, significantly reducing its moisture buffering capacity and creating well-developed capillary water loss channels. In terms of construction, the thin-layer paving process requires 5-20 minutes of open exposure, significantly increasing the convective contact area between the RCC and air; the vibratory compaction process disrupts the surface water film structure, further exacerbating the risk of moisture evaporation. This places higher demands on moisture content control during RCC construction. The dynamic evolution of internal temperature, humidity, resistivity, and pH value of RCC during the construction period is a core physical field indicator reflecting its hydration process and microstructure development. Accurately grasping the evolution law of these indicators and predicting the later performance has become a core requirement for the quality control of RCC projects under extreme environments.

[0003] However, existing technologies still have significant shortcomings in RCC performance control under strong wind construction environments: (1) The prediction model is highly general, but not adaptable to strong winds: Most existing prediction models are simple adaptations of general time series models. They do not quantify the nonlinear coupling relationship between strong winds and RCC physical fields, making it difficult to accurately capture specific scenarios such as sudden changes in physical field gradients caused by strong winds. The prediction accuracy drops significantly under extreme wind conditions.

[0004] (2) Passive hardware acquisition and lack of collaborative intelligence: Traditional hardware can only achieve fixed frequency data acquisition and basic preprocessing. It cannot adjust the acquisition strategy according to the dynamic changes of strong wind. Moreover, the sensor is easily affected by strong wind and mortar blockage, has weak durability, and relies on manual maintenance.

[0005] (3) The early warning mechanism is static and lacks causal tracing: existing early warnings are mostly based on fixed thresholds or adjustable parameters. They can only determine whether the standard is exceeded, but cannot predict the trend of performance deterioration or locate the root cause of the risk, resulting in insufficient targeting of control measures.

[0006] (4) The control process is semi-closed and lacks adaptive capability: The existing system has not formed a full-process adaptive optimization of "collection-prediction-early warning-control-feedback". The control measures rely on manual decision-making and cannot be dynamically iterated and optimized according to real-time working conditions. The ability to adapt to the dynamic changes of strong wind environment is limited. Summary of the Invention

[0007] This invention addresses the problems of complex RCC physical field evolution and strong wind coupling under strong wind construction conditions, lack of intelligent adaptation in hardware acquisition, inability to trace the root cause in early warning, and insufficient adaptive control processes. It provides an intelligent prediction and early warning system for roller-compacted concrete performance adapted to strong wind environments, comprising: An edge-intelligent collaborative monitoring layer is deployed in the roller-compacted concrete construction area to collect and preprocess multi-physics parameters of roller-compacted concrete, and dynamically optimize the collection strategy according to strong wind conditions. A dedicated algorithm processing layer for strong winds is connected to an edge intelligent collaborative monitoring layer. Based on multiple physical field parameters, it predicts the evolution trend of the physical field of roller-compacted concrete through a strong wind coupled spatiotemporal prediction model, and evaluates the later performance and risk root causes through a causal mapping mechanism. The causal dynamic early warning application layer is connected to the strong wind-specific algorithm processing layer, supports data visualization, triggers multi-level early warnings based on predicted evolution trends and risk assessment results, generates early warning reports containing the root causes of risks, and pushes adaptive control solutions.

[0008] Furthermore, the edge intelligent collaborative monitoring layer includes: The probe-type sensor array is equidistantly arranged along the thickness direction of the roller-compacted concrete, including the surface, middle, and bottom layers, as well as the horizontal direction. It integrates temperature, humidity, resistivity, pH, and miniature pressure sensors. The probes use titanium alloy electrodes, and the encapsulation shell has ventilation and water-permeable holes and a built-in miniature air pump. The edge intelligent collaborative node connects to the sensor array and has a built-in data preprocessing unit, outlier removal module, collaborative decision-making module, and equipment self-sustaining unit. The equipment self-sustaining unit triggers backflushing cleaning based on the pressure data from the micro pressure sensor. The dual-mode anti-interference collaborative transmission network adopts the LoRa-WiFi dual-mode wireless self-organizing network transmission module, which is suitable for inter-node collaborative communication, long-distance data transmission, and short-distance terminal interaction; a new dual-path cross-checking algorithm is added, which automatically switches to the backup link and corrects data deviation when strong winds cause signal interference. The working condition-physical field coupling acquisition unit integrates a wind speed sensor, an ambient temperature and humidity sensor, and a wind speed change rate calculation module to simultaneously acquire real-time wind force level and ambient temperature and humidity data of the construction environment. The intelligent data acquisition strategy optimization engine dynamically adjusts the data acquisition strategy based on wind speed level and physical field conditions.

[0009] Furthermore, the data preprocessing unit in the edge intelligent collaborative node is used for data normalization, noise reduction, and missing value completion; The outlier removal module sets a dynamic threshold based on the normal fluctuation range of the physical field under strong wind conditions to remove outlier data. The collaborative decision-making module builds an edge node collaborative network based on a distributed federated learning framework. Each node only encrypts and shares data features and model parameters without leaking the original data. Through a preset voting mechanism, it makes collaborative judgments and dynamically optimizes the acquisition frequency and triggering timing of sensors across the entire area by combining on-site wind speed and concrete physical field conditions.

[0010] Furthermore, the dedicated algorithm processing layer for strong winds includes: The strong wind-physical field coupled time series data fusion module receives temperature and humidity, resistivity, pH value, pressure data at various layers, as well as wind speed, wind speed change rate, and ambient temperature and humidity data. It constructs a five-dimensional time series dataset that includes wind speed, wind speed change rate, layer, time, and physical field, and highlights the coupling effect of strong wind through dynamic weight allocation. A strong wind coupled spatiotemporal prediction model, including a wind speed gradient coupling module, a gradient mutation adaptive submodule, and a layer-differentiated attention mechanism, is used to predict the evolution trend of the physical field of roller-compacted concrete by iterative optimization using the Adam optimizer with five-dimensional time series data as the training set. The physics-performance mapping module, based on test data under strong wind conditions, uses a Bayesian network to construct a causal graph of "strong wind - construction parameters - physics field - post-performance". It quantifies the causal relationship between various factors and strength, crack resistance, and durability, and establishes a multi-dimensional performance correlation model. It inputs physics field prediction data and outputs quantitative indicators such as the probability of strength compliance, cracking risk coefficient, and durability score, while also outputting the contribution percentage of each factor. The physics field prediction data specifically includes temperature, humidity, resistivity, pH value, and pressure along the concrete thickness direction for the surface, middle, and bottom layers. The causal inference-based dynamic early warning threshold module generates a dynamic threshold curve based on historical similar working condition data, and uses a preset percentage of the dynamic threshold as a trend warning line. It also calculates a comprehensive risk index by combining the probability of strength compliance, crack risk coefficient, and the causal contribution ratio of durability score.

[0011] Furthermore, in the strong wind coupled special spatiotemporal prediction model, the wind speed gradient coupling unit quantifies the nonlinear influence of strong wind on the physical field at different layers through a dynamic weight matrix, and establishes a mathematical relationship between wind speed and moisture evaporation, temperature and humidity conduction. The gradient mutation adaptive submodule sets up an LSTM decoder and a sliding window to monitor the first derivative of the physical field parameters. When the rate of change of the first derivative of the physical field parameters reaches a preset threshold, it activates local feature reinforcement learning. The layer-differentiated attention mechanism configures multiple attention heads at the surface, middle and bottom layers to capture the evolutionary specificity of each layer.

[0012] Furthermore, the causal dynamic early warning application layer includes: The multidimensional evolution trend display module is used to display real-time data of the physical field, future predicted evolution curves and subsequent performance prediction trends, and is configured with a strong wind impact analysis panel to quantify the contribution ratio of wind speed to each physical field index. The causal traceability multi-level early warning module triggers corresponding early warning signals based on the risk level and generates an early warning report that includes the risk level, evolution trend, root cause, scope of impact, and targeted control recommendations. The root cause is presented in the form of the contribution ratio of each factor. The adaptive control decision engine automatically matches the optimal control scheme library based on causal inference results, evaluates the effectiveness of measures through real-time data feedback, and dynamically iterates and optimizes control parameters. The system's self-optimization module is used to continuously collect real-time data and control effect data, periodically update the causal graph coefficients and dynamic threshold curves, and automatically generate construction parameter optimization suggestions. The causal graph coefficients are the core quantitative indicators that quantify the strength of causal relationships and contribution ratios among factors at each level in the "strong wind-construction parameters-physical field-subsequent performance" causal graph constructed by Bayesian network. The full lifecycle data management module stores monitoring data, forecast data, early warning data, operating condition data, and the implementation effects of control measures in a database, and supports querying and exporting by time range, level, and data type.

[0013] Furthermore, based on causal inference results, the automatic matching of the optimal control scheme library specifically includes: When the root cause of the risk is excessive wind speed, a control plan should be implemented, including adding windproof barriers, surface spraying for moisturizing, and shortening the paving exposure time. When the root cause of the risk is the accumulation of hydration heat, then we will push for adjustments to the water cooling flow rate, surface ventilation, and optimized mix ratio control measures. When the root cause of the risk is insufficient maintenance, a control plan should be implemented to increase the frequency of watering the surface of the roller-compacted concrete and cover it with heat-insulating and moisture-retaining cotton blankets.

[0014] A method for intelligent prediction and early warning of roller-compacted concrete performance adapted to strong wind environments is applied to an intelligent prediction and early warning system for roller-compacted concrete performance adapted to strong wind environments, comprising the following steps: S1. Pre-construction preparation: During the construction preparation phase, the sensor array, edge nodes, transmission modules, and operating condition acquisition units are installed, debugged, and their parameters are initialized. S2. Dynamic acquisition and preprocessing: After the pouring is completed, the system is started. The edge intelligent collaborative monitoring layer dynamically adjusts the acquisition strategy according to the wind speed and physical field state to complete data preprocessing and anti-interference transmission. S3. Coupled Prediction and Causal Analysis: The strong wind-specific algorithm processing layer receives data, constructs a five-dimensional time series dataset, predicts the evolution of the physical field through a strong wind-coupled dedicated spatiotemporal prediction model, and evaluates the later performance and risk root causes based on the physical field-performance causal mapping module. S4. Dynamic Early Warning and Control: The causal dynamic early warning application layer triggers corresponding level early warnings based on prediction and evaluation results, pushes root cause analysis and control solutions, and receives execution feedback. S5. Adaptive iterative optimization: The system continuously collects real-time data and control effect data, iteratively updates model parameters, dynamic thresholds and control schemes, forming a closed-loop optimization throughout the entire process.

[0015] Furthermore, the construction of the five-dimensional time series dataset in step S3 specifically includes: For the data received by the strong wind-specific algorithm processing layer, Z-score normalization is used to eliminate the influence of dimensions, and a five-dimensional time-series dataset containing wind speed, wind speed change rate, layer, time, and physical field is constructed. By dynamically assigning weights, the coupling effect of strong winds is highlighted, and the effectiveness of the dataset is optimized. Input the five-dimensional time series dataset into the strong wind coupled dedicated spatiotemporal prediction model to predict the evolution of the physical field.

[0016] Compared with the prior art, the present invention achieves the following technical effects: 1. Significantly improved prediction accuracy: This invention constructs a dedicated spatiotemporal prediction model coupled with strong winds, embedding a wind speed gradient coupling module and a gradient mutation adaptive sub-mechanism. This enables it to accurately capture specific scenarios such as abrupt changes in the physical field gradient caused by strong winds. Compared to existing general time series models, this invention improves prediction accuracy by more than 15% under extreme wind conditions, accurately capturing abrupt changes in the physical field gradient caused by strong winds, and effectively solving the core problem of inaccurate prediction of RCC physical field evolution under extreme wind conditions.

[0017] 2. Significantly enhanced data acquisition efficiency and reliability: This invention employs an edge intelligent collaborative monitoring network, which can dynamically optimize the acquisition strategy based on strong wind conditions and physical field status, avoiding invalid acquisition and data loss. The sensor's built-in micro-pump enables a back-blowing self-cleaning function, reducing manual maintenance costs by 80%. Combined with LoRa-WiFi dual-mode anti-interference transmission and a dual-path cross-validation algorithm, the data transmission success rate can reach over 99%, solving the problems of passive hardware acquisition, weak durability, and reliance on manual maintenance in existing technologies.

[0018] 3. Breakthroughs in early warning foresight and root cause localization capabilities: This invention constructs a causal inference-based dynamic early warning mechanism. By generating dynamic threshold curves through temporal transfer learning, quality risks can be predicted 2-4 hours in advance. At the same time, the causal graph constructed based on Bayesian networks can quantify the contribution ratio of each factor to performance, accurately locate the root cause of risks, and improve the pertinence of control measures by 60%, thus overcoming the shortcomings of traditional early warning and blind prevention and control.

[0019] 4. Full-process adaptive control achieves closed-loop optimization. This invention forms a full-process adaptive closed loop of "collection-prediction-early warning-control-feedback". Through the adaptive control decision engine, the optimal control scheme is automatically matched and the parameters are dynamically iterated and optimized based on real-time data feedback. This significantly reduces the dependence on manual intervention and can flexibly adapt to the dynamic changes of strong wind environment.

[0020] Through the comprehensive application of the above-mentioned technical means, this invention increases the quality compliance rate of RCC projects to over 98%, reduces the cost of addressing quality hazards by 70%, and provides a solid guarantee for the safe construction and long-term operation and maintenance of RCC projects under extreme construction environments such as strong winds. Attached Figure Description

[0021] For ease of explanation, the present invention will be described in detail below with reference to specific embodiments and accompanying drawings.

[0022] Figure 1 A schematic diagram of an intelligent prediction and early warning system for roller-compacted concrete performance adapted to strong wind environments provided by the present invention; Figure 2 A flowchart of a method for intelligent prediction and early warning of roller-compacted concrete performance adapted to strong wind environments provided by the present invention; Figure 3 Platform interface system parameter settings to match the RCC performance intelligent prediction and early warning system in the embodiments of the present invention; Figure 4 This is a platform interface real-time data dashboard that matches the RCC performance intelligent prediction and early warning system in the embodiments of the present invention; Figure 5 This is a demonstration of the platform interface evolution trend that matches the RCC performance intelligent prediction and early warning system in the embodiments of the present invention; Figure 6 This is a platform interface with multi-level early warning information that matches the RCC performance intelligent prediction and early warning system in the embodiments of the present invention; Figure 7 Recommendations for platform interface management and control to match the RCC performance intelligent prediction and early warning system in this embodiment of the invention; Figure 8 To provide a platform interface for data querying and tracing that matches the RCC performance intelligent prediction and early warning system in this embodiment of the invention. Detailed Implementation

[0023] The following are specific embodiments of the present invention, described in conjunction with the accompanying drawings, to further illustrate the technical solutions of the present invention. However, the present invention is not limited to these embodiments. Specific details, such as particular configurations, are provided in the following description merely to aid in a comprehensive understanding of the embodiments of the present invention. Therefore, those skilled in the art should understand that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present invention.

[0024] It should be noted that, unless otherwise specified, the embodiments and features described in this invention can be combined with each other.

[0025] Example 1: like Figure 1 As shown, it discloses an intelligent prediction and early warning system for the performance of roller-compacted concrete that is adapted to strong wind environments, specifically including: an edge intelligent collaborative monitoring layer, a strong wind-specific algorithm processing layer, and a causal dynamic early warning application layer.

[0026] An edge-intelligent collaborative monitoring layer is deployed in the roller-compacted concrete construction area. Based on a dynamic optimization acquisition strategy under strong wind conditions, it achieves accurate and efficient acquisition and preprocessing of RCC multi-physics parameters, providing high-quality data support for the algorithm layer. Specifically, this includes: The probe-type sensor array is equidistantly arranged along the RCC thickness direction (top, middle, and bottom layers) and horizontally. The spacing between the surface layer sensors can be set to 50cm, the spacing between the middle and bottom layers can be set to 80cm, and one sensor can be arranged every 1.5m in the horizontal direction, forming a layered and differentiated sensor array. The sensor operates on a DC power supply of 4.5~30V and integrates temperature (-40~80℃, 0.1℃ resolution), humidity (0~100%RH, 0.1%RH resolution), resistivity (0~20000μS / cm, 1μS / cm resolution), pH value (3-14, 0.1 resolution), and a newly added miniature pressure sensor (0~10kPa, 0.01kPa resolution). It has a built-in temperature compensation sensor (0~50℃) and an IP68 waterproof rating. The probe uses antifreeze, vibration-resistant, and corrosion-resistant titanium alloy electrodes, and the encapsulation shell is made of polytetrafluoroethylene material with 0.08mm breathable and water-permeable holes. It also has a built-in miniature air pump. The output signal follows the LoRa-WiFi dual-mode protocol, and a new anomaly prediction microprocessor with a preset physical field anomaly prediction rule library is added.

[0027] Among them, the preset physical field anomaly prediction rule library is a set of anomaly judgment rules pre-installed in the sensor microprocessor. It is based on the normal fluctuation range of the physical field of roller-compacted concrete under strong wind conditions, industry standards and engineering measured anomaly characteristics. With index values, change rates and gradient thresholds as the core, it covers the differentiated judgment standards of core parameters such as temperature and humidity and resistivity and derived indicators (matching wind speed level and concrete layer). The sensor can identify physical field anomalies through its real-time self-test, providing a judgment basis for subsequent anomaly value elimination and acquisition strategy adjustment.

[0028] The edge intelligent collaborative node utilizes low-power edge computing nodes connected to a probe-type sensor array. It integrates: a data preprocessing unit, employing an STM32H743 main control chip, for data normalization, noise reduction, and missing value completion; an outlier removal module, setting dynamic thresholds based on the normal fluctuation range of the physical field under strong wind conditions to remove abnormal data; a collaborative decision-making module, building an edge node collaborative network based on a distributed federated learning framework. Each node only encrypts and shares data features and model parameters, without leaking raw data. Through a preset voting mechanism, it collaboratively judges and, combined with on-site wind speed and concrete physical field conditions, jointly decides on acquisition strategy parameters such as sensor acquisition frequency and additional sampling triggering timing for each layer, achieving unified dynamic optimization of the acquisition strategy across the entire monitoring area and avoiding the problem of unreasonable strategies caused by local data deviations from a single node; and a self-sustaining unit that receives sensor pressure data and triggers a micro-pump for backflushing cleaning when the pressure is ≥5kPa to prevent mortar blockage. The edge node is solar-powered, with standby power consumption ≤50mW, supporting 72 hours of continuous operation without sunlight. The determination of the normal fluctuation range is based on a special test of the physical field of roller-compacted concrete covering the full wind speed range of 3-12. The test obtains the natural fluctuation range of core indicators such as temperature, humidity, resistivity, and pH value under different strong winds to determine the basic reference range. Then, it is combined with the construction history monitoring data of similar strong wind projects and corrected according to different layers and construction stages. At the same time, it follows the upper and lower boundaries of the indicators in the "Specification for Construction of Hydraulic Roller-Compacted Concrete" DL / T5112-2021 and will also be dynamically fine-tuned according to the real-time wind speed level on site (such as extreme wind conditions) to adapt to the actual changes of the physical field under strong winds.

[0029] The dual-mode anti-interference collaborative transmission network adopts a LoRa-WiFi dual-mode wireless self-organizing network transmission module. The LoRa mode is set to the 433MHz frequency band, with a transmission distance of 1.5km and a rate of 9600bps, which can be used for inter-node collaborative communication and long-distance data transmission. The WiFi mode is set to the 2.4GHz frequency band with a rate of 150Mbps, which can be used for short-range terminal interaction. A new dual-path cross-checking algorithm is added to automatically switch to the backup link and correct data deviation when strong winds cause signal interference. The dual-path cross-validation algorithm is a dedicated verification mechanism designed to address the issues of interference and data distortion in wireless transmission under strong wind construction conditions. Its core mechanism uses a dual-mode parallel transmission of the same set of monitoring data via a primary link in LoRa mode and a backup link in WiFi mode. A dual-verification logic is constructed in the edge intelligent node: First, it performs timestamp alignment and numerical deviation analysis on the data transmitted through the two links, setting a reasonable deviation threshold, which can be ≤0.5%. When strong winds cause signal attenuation or interference leading to a single link's data exceeding the threshold or transmission delay exceeding the timeout, it automatically determines that the link's data is invalid and switches to the backup link. The transmission delay timeout can be set to >3 seconds. Second, based on the temporal continuity characteristics of concrete physical field parameters (such as temperature and humidity not changing abruptly in a short time), it uses a sliding window (3 consecutive data points) to smooth and correct the effective link data, eliminating abnormal jump values ​​caused by instantaneous interference. Simultaneously, it combines a transmission error model under similar historical working conditions to compensate for the accuracy of the corrected data, ultimately outputting highly reliable, low-deviation monitoring data to ensure the accuracy of subsequent prediction models and early warning mechanisms.

[0030] The working condition-physical field coupled acquisition unit integrates a wind speed sensor, an ambient temperature and humidity sensor, and a wind speed change rate calculation module. It simultaneously collects real-time wind force level and ambient temperature and humidity data at the construction site, which serve as auxiliary features for time series prediction in the algorithm layer, thereby improving the adaptability of the prediction model to strong wind conditions.

[0031] The intelligent data acquisition strategy optimization engine dynamically adjusts the acquisition strategy based on wind speed level and physical field conditions, specifically including the following acquisition strategies: (1) Under normal operating conditions, i.e., wind speed < level 6 and no abnormal physical field, the sampling frequency is set to 1 time / 10min; (2) Wind condition changes, i.e., under the condition of wind speed ≤ level 6 < level 8, the sampling frequency of the surface sensor is increased to 1 time / 5min through node cluster voting, while the middle and bottom layers are kept at 1 time / 10min. ≥ 60% of the node cluster votes indicate that the node cluster agrees. (3) In extreme wind conditions, i.e., when the wind speed is ≥ level 8, the sampling frequency is set to 1 time / 3min for the surface layer, 1 time / 5min for the middle layer, and 1 time / 10min for the bottom layer. (4) Physical field anomalies, i.e., when any condition in the rule base is met, trigger additional sampling, with a maximum sampling frequency of 2 times / 10min.

[0032] A dedicated algorithm processing layer for strong winds connects with an edge intelligent collaborative monitoring layer. Based on multi-physics parameters, it predicts the evolution trend of the physical field of roller-compacted concrete using a strong wind-coupled spatiotemporal prediction model. Furthermore, it evaluates subsequent performance and root causes of risks through a causal mapping mechanism. Specifically, this includes: The strong wind-physical field coupled time-series data fusion module receives temperature, humidity, resistivity, pH, and pressure data from various layers transmitted by the edge intelligent collaborative monitoring layer, as well as auxiliary data such as wind speed, wind speed change rate, and ambient temperature and humidity. Z-score standardization is used to eliminate the influence of dimensions. A five-dimensional time-series dataset is constructed, including wind speed, wind speed change rate, layer, time, and physical field. Dynamic weight allocation highlights the coupling effect of strong winds and optimizes the effectiveness of the dataset. Specifically, the dynamic weight allocation can be set to 0.28-0.20 for core physical field indicators when the wind speed is <6, and to 0.35 for wind speed ≥6.

[0033] Strong wind coupled dedicated spatiotemporal prediction models include: The wind speed gradient coupling module quantifies the nonlinear impact of strong winds on the physical fields of different layers through a dynamic weight matrix, establishing a mathematical relationship between wind speed and moisture evaporation, temperature and humidity conduction, such as surface humidity evaporation rate = 0.02 × wind speed + 0.005 × wind speed change rate - 0.01. The dynamic weight matrix of the wind speed gradient coupling module is a multi-dimensional matrix containing wind speed level, concrete layer, and physical field indicators (temperature, humidity, resistivity, pH value, etc.). As the wind speed level increases, the weight of the surface layer increases, while the weight of the middle and lower layers decreases. For every 1-level increase in wind speed, the weight of the surface layer increases by 0.08~0.12, the middle layer by 0.03~0.05, and the weight of the lower layer remains basically stable. The weights of indicators directly affected by strong winds are specifically strengthened.

[0034] The gradient mutation adaptive submodule sets up an LSTM decoder and a sliding window to monitor the first derivative of the physical field parameters. When the rate of change of the first derivative of the physical field parameters reaches a preset threshold, local feature reinforcement learning is activated. The sliding window is set to three time steps, and the preset threshold is set to the rate of change of the first derivative of the physical field parameters ≥ 10% / 10min.

[0035] The layer-specific attention mechanism is configured with four attention heads in the surface layer to capture the evolutionary trend of water loss; three attention heads in the middle layer to capture the evolutionary trend of heat of hydration accumulation; and two attention heads in the bottom layer to capture the evolutionary trend of humidity retention, thereby capturing the evolutionary specificity of each layer.

[0036] The strong wind coupled spatiotemporal prediction model uses five-dimensional time-series data as the training set, employs the Adam optimizer, iterates for 100 rounds until the MSE ≤ 0.003, and embeds an adaptive learning mechanism to fine-tune parameters every 24 hours using real-time data to avoid overfitting. The Adam optimizer can be configured with a learning rate of 0.001 and a decay coefficient of 0.9.

[0037] The physics-performance mapping module, based on extensive test data covering wind speeds of 3-12 levels and different curing ages under strong wind conditions, uses a Bayesian network to construct a causal graph of "strong wind-construction parameters-physics field-post-performance". It quantifies the causal relationship between 12 categories of factors, including wind parameters (wind speed, wind speed change rate, wind field distribution), construction parameters (curing frequency, paving exposure time, mix proportion, water cooling flow rate, compaction process parameters), and concrete physical field parameters (temperature and humidity gradients at each layer, resistivity change rate, pH value, internal pressure distribution), and strength, crack resistance, and durability. For example, the causal coefficient of wind speed on surface cracking is 0.72. A multi-dimensional performance correlation model is established, taking the physical field prediction data as input and outputting quantitative indicators such as the probability of strength compliance at 7 days and 28 days, crack risk coefficient, and durability score, while also outputting the contribution percentage of each factor. Based on common sense, the contribution percentages for the probability of strength compliance, crack risk coefficient, and durability score are set at 0.4 and 0.2 respectively.

[0038] The quantification of causal relationship strength is based on test data of strong wind conditions with wind speeds of 3-12 levels and different curing ages. After normalization, a conditional probability table of each factor and concrete performance is constructed. The causal relationship strength is quantified by causal coefficients in the range of 0 to 1 through conditional probability calculation and causal inference using Bayesian networks. The closer the coefficient is to 1, the more significant the impact. When multiple factors are coupled, the comprehensive relationship strength is calculated by joint probability calculation. At the same time, direct and indirect relationships are distinguished and the corresponding coefficients are labeled.

[0039] The contribution ratio of each factor is first determined by setting the basic weights of the performance dimensions according to common sense in engineering management, such as strength 0.4, cracking 0.4, and durability 0.2. Then, the proportion coefficient of the single factor causality coefficient to the sum of the causality coefficients of all related factors under its corresponding performance index is calculated. This coefficient is multiplied by the corresponding basic weight of the performance to obtain the comprehensive contribution ratio of the single factor. When multiple factors are coupled, the coupling contribution ratio of each factor is split according to the ratio of the coupling joint causality coefficient to ensure that the sum of the proportions of all factors is 1.

[0040] A causal inference-based dynamic early warning threshold module, combining the RCC engineering quality acceptance standards under strong wind construction conditions (DL / T5112-2021 "Code for Construction of Roller-Compacted Concrete in Hydraulic Engineering" and GB50204-2015 "Code for Acceptance of Construction Quality of Concrete Structures"), a large amount of RCC performance test data under strong wind conditions, and the actual quality control needs of the project, constructs a dynamic evolution early warning system. Based on historical similar working condition data, a dynamic threshold curve is generated through time-series transfer learning. For example, when the wind speed increases from level 5 to level 7, the temperature and humidity gradient early warning threshold is lowered from 25℃ / m to 22℃ / m; and 80% of the dynamic threshold is used as the trend early warning line to predict risks 2-4 hours in advance; cracking risk, strength failure risk, and durability risk are selected as the judgment index (0~100) for the comprehensive risk level, divided into four levels: no risk (0~20 points), general risk (21~40 points), high risk (41~70 points), and extremely high risk (71~100 points).

[0041] The causal dynamic early warning application layer connects with the strong wind-specific algorithm processing layer, supports data visualization, triggers multi-level early warnings based on predicted evolution trends and risk assessment results, generates early warning reports containing the root causes of risks, and pushes adaptive control solutions, specifically including: The multi-dimensional evolution trend display module displays real-time physical field data, evolution curves for the next 12-72 hours, and subsequent performance prediction trends through line graphs, heat maps, etc.; it supports comparing evolution differences at different wind speeds and different layers; and it adds a strong wind impact analysis panel to quantify the contribution ratio of wind speed to various physical field indicators.

[0042] The causal-tracing multi-level early warning module triggers corresponding early warning signals based on the risk level. No risk triggers a green alert, moderate risk triggers a yellow alert, higher risk triggers an orange audible and visual alarm, and extremely high risk triggers a red audible and visual alarm and SMS notification. Simultaneously, it generates an early warning report including the risk level, evolution trend, root cause, scope of impact, and targeted control recommendations. The root cause is presented as the contribution percentage of each factor; for example, wind speeds of level 7 or higher and insufficient maintenance frequency contribute 65% and 35% respectively.

[0043] The full lifecycle data management module uses an SQL Server 2019 database to store monitoring data, forecast data, early warning data, operating condition data, and the implementation effect of control measures through dual local and cloud backups, with a storage time of 5 years or more; it supports querying and exporting by time range, level, and data type, providing data support for model iteration and cross-project reference.

[0044] The adaptive control decision engine, based on causal inference, automatically matches the optimal control scheme library. If the root cause of the risk is excessive wind speed, it pushes control schemes such as adding windbreaks, surface spraying for moisture retention, and shortening paving exposure time. If the root cause is hydration heat accumulation, it pushes control schemes such as adjusting water cooling flow, surface ventilation, and optimizing mix proportions. The optimized mix proportions further increase the fly ash content while ensuring performance requirements, thereby reducing the temperature by decreasing hydration heat release. If the root cause is insufficient curing, it pushes control schemes such as increasing watering frequency and covering with heat-insulating and moisture-retaining blankets. The effectiveness of the measures is evaluated through real-time data feedback, and control parameters are dynamically iterated and optimized. A setting can be made to consider a measure effective when the humidity recovery rate is ≥10%.

[0045] The system's self-optimization module continuously collects real-time data and control effect data, updating the causal graph coefficients and dynamic threshold curves every 24 hours based on the daily data; it integrates cross-project data every 30 days to optimize the generalization ability of the prediction model; and it automatically generates optimization suggestions for construction parameters under strong wind conditions, providing a reference for subsequent projects. The causal graph coefficients are core quantitative indicators that quantify the strength of causal relationships and the contribution ratio between factors at each level in the "strong wind-construction parameters-physical field-later performance" causal graph constructed by the Bayesian network. Specifically, they cover three core types of coefficients: strong wind-related factors such as wind speed and wind speed change rate; construction parameters such as curing frequency, mix proportion, and paving process; and physical field indicators of each concrete layer, such as temperature, humidity, resistivity, and pH value, and their correlation coefficients, weighting coefficients, and contribution ratio coefficients with the later strength, crack resistance, and durability of roller-compacted concrete. These coefficients intuitively reflect the degree of influence of single or multiple factors coupled on concrete performance indicators.

[0046] Example 2: like Figure 2 As shown, this invention provides an intelligent prediction and early warning method for the performance of roller-compacted concrete adapted to strong wind environments, achieving a closed-loop process. Specifically, it includes the following steps: S1. During the construction preparation stage, the installation and commissioning of the sensor array, edge nodes, transmission modules and working condition acquisition units are completed three days before the pouring, and the sensors are calibrated to ensure that the error is ≤ ±0.5%, and parameters such as the initial acquisition frequency and dynamic threshold generation rules are set. S2. Dynamic acquisition and preprocessing: After the pouring is completed, the system is started. The edge intelligent collaborative monitoring layer dynamically adjusts the acquisition strategy according to the wind speed and physical field state to complete data preprocessing and anti-interference transmission. S3. Coupled Prediction and Causal Analysis: The strong wind-specific algorithm processing layer receives data, constructs a five-dimensional time series dataset, predicts the evolution of the physical field through a strong wind-coupled dedicated spatiotemporal prediction model, and evaluates the later performance and risk root causes based on the physical field-performance causal mapping module. S4. Dynamic Early Warning and Control: The causal dynamic early warning application layer triggers corresponding level early warnings based on prediction and evaluation results, pushes root cause analysis and control solutions, and receives execution feedback. S5. Adaptive iterative optimization: The system continuously collects real-time data and control effect data, iteratively updates model parameters, dynamic thresholds and control schemes, forming a closed-loop optimization throughout the entire process.

[0047] Example 3: This embodiment, based on Embodiments 1 and 2, combines the construction scenario of a C30 strength grade RCC dam in a high-altitude, windy region (normal wind force 3-8, extreme force 10, design age 28 days) to illustrate the implementation details of the invention, ensuring that the construction quality meets the requirements of the "Specification for Construction of Hydraulic Roller-Compacted Concrete" DL / T5112-2021 and the "Specification for Acceptance of Construction Quality of Concrete Structures" GB50204-2015.

[0048] The system is deployed in a three-layer architecture: an edge intelligent collaborative monitoring layer, a strong wind-specific algorithm processing layer, and a causal dynamic early warning application layer. These layers are located at the dam construction site and the monitoring center. The edge intelligent collaborative monitoring layer equipment is designed to withstand strong winds and low temperatures and is deployed in the pouring area and at high points on the construction site. The strong wind-specific algorithm processing layer is deployed on an industrial computer (Intel Core i7-12700H, 16GB RAM, 512GB hard drive) in the monitoring center. The causal dynamic early warning application layer consists of fixed terminals in the monitoring center and mobile terminals on-site, compatible with Android and iOS systems, enabling multi-terminal collaborative management.

[0049] Edge intelligent collaborative monitoring layer deployment The edge intelligent collaborative monitoring layer is a strong wind-adaptive intelligent acquisition network, which includes LoRa-WiFi probe sensors, edge intelligent collaborative nodes, dual-mode anti-interference collaborative transmission network, working condition-physical field coupling acquisition unit and intelligent acquisition strategy optimization engine. The core of it realizes accurate acquisition, preprocessing, anti-interference transmission and dynamic optimization of acquisition strategy of RCC multi-physical field parameters under strong wind conditions.

[0050] The probe-type sensor was deployed and debugged. The sensor is a strong wind-enhanced multi-parameter probe-type sensor, specifically the WSN-RW-5P model (including pressure monitoring), powered by a 12V solar battery. The sensor integrates the following parameter acquisition functions: temperature (-40~80℃, 0.1℃ resolution), humidity (0~100%RH, 0.1%RH resolution), resistivity (0~20000μS / cm, 1μS / cm resolution), pH value (3-14, 0.1 resolution), and pressure (0~10kPa, 0.01kPa resolution). Specific sensor parameter selection is as follows... Figure 3As shown. Sensors are deployed along the surface, middle, and bottom layers of the dam. The surface layer has three sensors spaced 50cm apart, while the middle and bottom layers each have three sensors spaced 80cm apart. One sensor is deployed every 1.5m horizontally, forming a 3×10 sensor array. The encapsulated shell has 0.08mm diameter vents for air and water permeability. The built-in micro air pump has a trigger pressure threshold of 5kPa. When the pressure reaches the threshold, it automatically triggers backflushing for cleaning to prevent mortar blockage.

[0051] The edge intelligent collaborative node configuration and connection uses the EC-2000 edge node, which connects to the sensor array via RS-485, and is equipped with a 100W solar power supply module and a 20000mAh battery. The edge node integrates the following functional modules: data preprocessing uses min-max normalization, 5th-order moving average denoising, and linear interpolation + moving average to complete missing values; the outlier removal threshold is set at temperature -10~60℃, humidity 30%~90%RH, resistivity 500~15000μS / cm, and pH 12~13.5, and data outside the range is removed; collaborative decision-making adopts a distributed federated learning framework, with data feature sharing among nodes, and a voting threshold of 60% to participate in the collaborative decision-making of the acquisition strategy; the device self-sustaining unit receives sensor pressure data, and triggers a micro air pump for backflushing cleaning when the pressure is ≥5kPa.

[0052] A dual-mode anti-interference collaborative transmission network was established, using the WT-3000-LW transmission module. The LoRa band operates at 433MHz, with a transmission distance of 1.5km and a speed of 9600bps. The WiFi band operates at 2.4GHz, with a speed of 150Mbps. A dual-path cross-validation algorithm was implemented, with a verification error threshold of ≤0.5%. If the threshold is exceeded, the system automatically switches to a backup link and corrects data deviations.

[0053] The working condition-physical field coupling acquisition unit is installed by installing FS-3000 wind speed sensors and TH-2000 environmental temperature and humidity sensors at the highest point of the dam. The acquisition frequency is synchronized with the sensors at 1 time / 10min and supports dynamic adjustment. It has a built-in wind speed change rate calculation module with a calculation window set to 3 time steps. It synchronously collects real-time wind force level, environmental temperature and humidity, and wind speed change rate data at the construction site as auxiliary features for time series prediction in the algorithm layer.

[0054] The intelligent data acquisition strategy optimization engine parameter settings are as follows: the initial acquisition frequency is set to 1 time / 10 minutes, and the dynamic adjustment strategy is as follows: Under normal operating conditions, i.e. wind speed < level 6 and no physical field anomalies, the sampling frequency is set to 1 time / 10min. When wind conditions change, i.e., wind speed ≤ level 6 < level 8, and wind speed 10.8~17.1m / s, the sampling frequency of the surface sensor is increased to once / 5min through node cluster voting, while the sampling frequency of the middle and bottom layers is kept at once / 10min. ≥60% of the node cluster vote indicates that the node cluster agrees. In extreme wind conditions, i.e. wind speed ≥ level 8 and wind speed ≥ 17.2 m / s, the sampling frequency is set to 1 time / 3 min for the surface layer, 1 time / 5 min for the middle layer, and 1 time / 10 min for the bottom layer. Physical field anomalies are triggered when any condition in the rule base is met, resulting in additional sampling, with a maximum sampling frequency of 2 times per 10 minutes.

[0055] The physical field anomaly rule library includes: humidity decrease rate ≥ 5% / 10min, temperature gradient ≥ 20℃ / m, resistivity change rate ≥ 8% / 10min.

[0056] Implementation and operation of the dedicated algorithm processing layer for strong winds The strong wind-specific algorithm processing layer is deployed on the industrial computer in the monitoring center. It is developed based on the TensorFlow and PyTorch frameworks and includes a strong wind-physical field coupled time series data fusion module, a strong wind coupled dedicated spatiotemporal prediction model, a physical field-performance causal mapping module, and a causal inference-type dynamic early warning threshold module. The core functions are multi-source data fusion, high-precision prediction of physical field evolution under strong wind coupling, performance correlation analysis, and dynamic early warning threshold generation.

[0057] The strong wind-physical field coupling time series data fusion module, developed based on the TensorFlow framework, receives data from sensor arrays and operational condition acquisition units, and uses Z-score normalization to eliminate dimensional errors. The five-dimensional time series dataset is set as follows: time step (10 min / step) × number of layers (3 layers) × number of points (30) × number of parameters (6 items). Finally, a dynamic weight allocation strategy is used to optimize the dataset. When the wind speed is < level 6, the weights of temperature 0.28, humidity 0.27, resistivity 0.25, and pH value 0.20 are assigned to highlight the core physical field indicators. When the wind speed is ≥ level 6, the wind speed weight is set to 0.20, the wind speed change rate is 0.15, the weights of the core physical field indicators are reduced accordingly, and the weight of the total wind speed-related features is increased to 0.35 to highlight the impact of strong wind coupling on the RCC physical field.

[0058] Training and inference of a dedicated spatiotemporal prediction model coupled with strong winds: Based on the PyTorch framework, a dedicated spatiotemporal prediction model coupled with strong winds is developed. The spatiotemporal Transformer encoder has 6 attention heads, configured differently according to the layer. Four attention heads are configured in the surface layer to focus on capturing the evolution trend of water loss; three attention heads are configured in the middle layer to focus on capturing the evolution trend of hydration heat accumulation; and two attention heads are configured in the bottom layer to focus on capturing the evolution trend of humidity retention, capturing the evolution specificity of each layer. The hidden layer dimension is 256. The LSTM decoder has 2 hidden layers (128 neurons / layer), embedding a gradient mutation adaptive submodule, and the sliding window size is set to 3 time steps. The model was trained using 4320 sets of five-dimensional time-series data from the 30 days prior to construction. The Adam optimizer was employed with a learning rate of 0.001 and a decay coefficient of 0.9, and the model underwent 100 iterations. Training was stopped when the mean squared error (MSE) ≤ 0.003. An adaptive learning mechanism was embedded, using 144 sets of data collected daily in real-time every 24 hours to fine-tune the model parameters and avoid overfitting. During the model inference phase, the fused five-dimensional time-series data was input, and the output was the evolution data of the physical field at each layer of the dam's RCC, which could be adjusted as needed to 72 hours over the next 24 hours. The prediction step size was 10 minutes, and the measured prediction accuracy reached 96.2%. When the rate of change of the first derivative of the physical field parameters was detected to be ≥ 10% / 10 minutes, the local feature reinforcement learning of the gradient mutation adaptive submodule was automatically activated to improve the prediction accuracy under scenarios of sudden gradient changes in the physical field caused by strong winds.

[0059] The physical field-performance causal mapping module was constructed and applied. Based on 100 sets of RCC test data covering strong wind conditions of level 3-12, a Bayesian network was used to construct a causal graph of "strong wind-construction parameters-physical field-post-performance". The causal correlation strength between 12 factors, including wind speed, maintenance frequency, and mix proportion, and RCC surface cracking, strength development, and durability was quantified. Specifically, the causal coefficients for wind speed and surface cracking were 0.72, maintenance frequency and surface cracking, and mix proportion and surface cracking, respectively, were 0.68 and 0.55. A multi-dimensional performance correlation model was trained using a multiple linear regression algorithm, setting a strength compliance probability ≥95% and a durability score ≥80 as the passing standard. The physical field prediction data output from the strong wind coupled dedicated spatiotemporal prediction model was input into this module, which automatically outputs quantitative performance indicators such as the RCC 7d and 28d strength compliance probability, cracking risk coefficient, and durability score, while also outputting the contribution percentage of each influencing factor to performance degradation / compliance. The contribution percentages for strength compliance probability, crack risk coefficient, and durability score are set at 0.4 and 0.2 respectively.

[0060] The causal inference-based dynamic early warning threshold module is configured to combine relevant specifications for hydraulic roller-compacted concrete construction, strong wind test data, and actual engineering quality control requirements to construct a dynamic evolution early warning system. Based on 50 sets of historical similar working condition data with a wind speed and construction parameter matching degree of ≥80%, a dynamic threshold curve is generated through time-series transfer learning (learning window set to 7 days). For example, when the wind speed increases from level 5 to level 7, the temperature and humidity gradient early warning threshold is lowered from 25℃ / m to 22℃ / m. Using 80% of the dynamic threshold as the trend early warning line, warnings are issued 2-4 hours in advance. Assess quality risks; select cracking risk, strength failure risk, and durability risk, and calculate a comprehensive risk index of 0-100 points based on the causal contribution ratio output by the physical field-performance causal mapping module. Divide the risk into four levels according to the index: no risk (0-20 points, strength compliance probability ≥95%), moderate risk (21-40 points, strength compliance probability 80%-94%), high risk (41-70 points, strength compliance probability 60%-79%), and extremely high risk (71-100 points, strength compliance probability <60%).

[0061] Implementation and deployment of the causal dynamic early warning application layer The causal dynamic early warning application layer is developed based on KingView 7.5 and Python PyQt5. It consists of a fixed terminal in the monitoring center and a mobile terminal in the field. As the core of the system's human-computer interaction, it realizes data visualization, causal traceability-based multi-level early warning, full lifecycle data management, adaptive control, and system self-optimization. The specific implementation functions and operational requirements are as follows: The multi-dimensional evolution trend display module is presented, such as Figure 4 , 5 As shown, the terminal interface displays a real-time data panel, a layered heat map, a line graph of physical field evolution, and a prediction curve of RCC performance in the later stages. It intuitively displays real-time physical field data of each layer and point of the dam's RCC, the evolution trend in the next 12-72 hours, and the probability of performance meeting standards in 7 days / 28 days. It supports the comparative display of physical field evolution data of different wind speed levels and different layers. A new "Strong Wind Impact Analysis Panel" has been added to quantify the contribution of real-time wind speed to each physical field index, providing construction personnel with intuitive data reference.

[0062] Triggered by a cause-and-effect tracing multi-level early warning module, such as Figure 6 , 7As shown, based on the comprehensive risk index output by the algorithm layer, corresponding warning signals are triggered at the terminal without additional manual intervention: No risk (0-20 points), the interface displays a green prompt "No quality risk, maintain routine monitoring"; General risk (21-40 points), the interface displays a yellow prompt "General risk, it is recommended to strengthen monitoring"; Higher risk (41-70 points), an orange audible and visual alarm is activated, prompting "Higher risk, take control measures immediately"; Extremely high risk (71-100 points), a red audible and visual alarm is activated and a text message is simultaneously sent to construction management personnel. When all warnings are triggered, a warning report is automatically generated, clearly indicating the risk level, the evolution trend of the physical field, the root cause of the risk and the contribution ratio of each cause (e.g., wind speed of 7.8 m / s and maintenance frequency of 2 hours / time, contribution ratios of 65% and 35% respectively), the scope of risk impact (e.g., the surface 3-5m area), and targeted control recommendations.

[0063] The full lifecycle data management module operates using an SQL Server 2019 database, establishing a dual-backup data storage system with local and cloud backups. Storage frequency is synchronized with data acquisition frequency, and the storage period is ≥5 years. Stored content includes RCC physical field monitoring data, algorithm prediction data, early warning data, construction site condition data, and data on the effectiveness of control measures. The database supports multi-dimensional queries based on time range, sensor number, data type, risk level, etc. Figure 8 As shown, the query results can be exported to Excel format, providing complete data support for model iteration optimization and cross-engineering technology reference.

[0064] The adaptive control decision engine matches and optimizes solutions. The control solution library contains 20 optimal solutions for typical scenarios. Based on the risk root causes and contribution ratios output by the causal traceability multi-level early warning module, it automatically matches the optimal control solution from the library and pushes it to the on-site mobile terminal. When wind speeds are ≥8, it pushes solutions such as adding windbreaks, surface spraying for moisturizing, shortening paving exposure time, and surface sampling once every 3 minutes. When hydration heat accumulates, it pushes solutions such as adjusting water cooling flow, surface ventilation, and optimizing mix ratios. The engine receives physical field monitoring data in real time after the implementation of control measures. The effectiveness of the measures is evaluated based on the recovery of physical field indicators to below the dynamic threshold. If the standard is not met, control parameters are automatically adjusted, such as increasing the watering frequency and adding layers of windbreaks, achieving dynamic iterative optimization of the control solution.

[0065] The system's self-optimization module runs continuously, interacting in real time with other modules. It continuously collects real-time monitoring data and data on the effectiveness of control measures. Every 24 hours, it updates the Bayesian network causal graph coefficients and dynamic threshold curves based on the latest data to ensure the accuracy of early warnings. Every 30 days, it integrates cross-project data from this project and similar RCC projects under strong wind conditions to train and optimize the dedicated spatiotemporal prediction model for strong wind coupling, improving the model's generalization ability. At the same time, it automatically generates a "Strong Wind Condition Construction Parameter Optimization Report" based on the full-cycle data, including suggestions on optimal aggregate gradation, cement dosage, curing frequency, paving process parameters, etc., providing technical reference for subsequent RCC project construction in strong wind environments.

[0066] The overall usage process of this system follows a closed loop of "construction preparation - dynamic data acquisition and preprocessing - coupled prediction and causal analysis - dynamic early warning and control - adaptive iterative optimization", with seamless connection between each link. The specific operation steps are as follows: S1. During the construction preparation phase, the installation, debugging, and networking of all equipment in the edge intelligent collaborative monitoring layer are completed 3 days before the RCC pouring of the dam; all sensors are calibrated to ensure that the calibration error is ≤±0.5%; the training and deployment of the algorithm layer model are completed in the industrial computer of the monitoring center, and basic parameters such as the initial acquisition frequency, dynamic threshold generation rules, and early warning level triggering conditions are set in the application layer terminal.

[0067] S2. Dynamic Acquisition and Preprocessing Stage: After the RCC is poured, the system is automatically started. The edge intelligent collaborative monitoring layer dynamically adjusts the acquisition frequency according to the real-time wind speed and RCC physical field status at the construction site to complete the acquisition of multi-physical field data and working condition data. The edge nodes perform preprocessing and outlier removal operations on the acquired data and transmit standardized high-quality data to the strong wind-specific algorithm processing layer through a dual-mode anti-interference transmission network.

[0068] S3. In the coupled prediction and causal analysis stage, the strong wind-specific algorithm processing layer receives data transmitted from the edge layer, constructs a five-dimensional time series dataset, and completes the prediction of the future evolution trend of the RCC physical field through the strong wind coupled dedicated spatiotemporal prediction model; based on the physical field-performance mapping module, it correlates the predicted physical field data with the later performance of RCC, quantifies performance indicators, and locates the root causes of risk and the contribution ratio of each cause; through the causal inference-type dynamic early warning threshold module, it calculates the comprehensive risk index and classifies the risk level.

[0069] S4. Dynamic early warning and control phase: The causal dynamic early warning application layer automatically triggers the corresponding level of early warning based on the comprehensive risk index, generates and pushes early warning reports and targeted control plans; construction personnel implement control measures based on the early warning reports, and the system collects physical field data in real time during the control process and feeds it back to the adaptive control decision engine.

[0070] In the S5 adaptive iterative optimization phase, the engine evaluates the effectiveness of control measures based on real-time data, and automatically adjusts control parameters if the targets are not met. The system's self-optimization module continuously collects real-time monitoring data and control effect data, updates the causal graph coefficients and dynamic threshold curves periodically, and optimizes the prediction model parameters. All data is synchronously stored in the full lifecycle data management module, forming an adaptive closed-loop optimization of "collection-prediction-early warning-control-feedback" without human intervention, continuously improving the system's adaptability and control accuracy in strong wind construction environments.

[0071] Those skilled in the art to which this application pertains may make various modifications or additions to the specific embodiments described, or adopt similar methods to replace them, without departing from the inventive concept of this application or exceeding the scope defined by the appended claims.

Claims

1. A smart predictive and early warning system for the performance of roller-compacted concrete adapted to strong wind environments, characterized in that, include: An edge-intelligent collaborative monitoring layer is deployed in the roller-compacted concrete construction area to collect and preprocess multi-physics parameters of roller-compacted concrete, and dynamically optimize the collection strategy according to strong wind conditions. A strong wind-specific algorithm processing layer is connected to the edge intelligent collaborative monitoring layer. Based on the multi-physics field parameters, it predicts the evolution trend of the physical field of roller-compacted concrete through a strong wind coupled spatiotemporal prediction model, and evaluates the later performance and risk root causes through a causal mapping mechanism. The strong wind-specific algorithm processing layer includes: The strong wind-physical field coupled time series data fusion module receives temperature and humidity, resistivity, pH value, pressure data at various layers, as well as wind speed, wind speed change rate, and ambient temperature and humidity data. It constructs a five-dimensional time series dataset that includes wind speed, wind speed change rate, layer, time, and physical field, and highlights the coupling effect of strong wind through dynamic weight allocation. A strong wind coupled spatiotemporal prediction model, including a wind speed gradient coupling module, a gradient mutation adaptive submodule, and a layer-differentiated attention mechanism, is used to predict the evolution trend of the physical field of roller-compacted concrete by iterative optimization using the Adam optimizer with five-dimensional time series data as the training set. The physics-performance mapping module, based on test data under strong wind conditions, uses a Bayesian network to construct a causal graph of "strong wind - construction parameters - physics field - post-performance". It quantifies the causal relationship between each factor and strength, crack resistance, and durability, and establishes a multi-dimensional performance correlation model. It inputs physics field prediction data and outputs quantitative indicators such as the probability of strength compliance, cracking risk coefficient, and durability score, while also outputting the contribution percentage of each factor. The physics field prediction data specifically includes temperature, humidity, resistivity, pH value, and pressure along the concrete thickness direction for the surface, middle, and bottom layers. The causal inference-based dynamic early warning threshold module generates a dynamic threshold curve based on historical similar working condition data, and uses a preset percentage of the dynamic threshold as a trend warning line. It also calculates a comprehensive risk index by combining the probability of strength compliance, crack risk coefficient, and the causal contribution ratio of durability score. The causal dynamic early warning application layer is connected to the strong wind-specific algorithm processing layer. It supports data visualization, triggers multi-level early warnings based on predicted evolution trends and risk assessment results, generates early warning reports containing the root causes of risks, and pushes adaptive control solutions.

2. The intelligent prediction and early warning system for roller-compacted concrete performance adapted to strong wind environments according to claim 1, characterized in that, The edge intelligent collaborative monitoring layer includes: The probe-type sensor array is equidistantly arranged along the thickness direction of the roller-compacted concrete, including the surface, middle, and bottom layers, as well as the horizontal direction. It integrates temperature, humidity, resistivity, pH, and miniature pressure sensors. The probes use titanium alloy electrodes, and the encapsulation shell has ventilation and water-permeable holes and a built-in miniature air pump. An edge intelligent collaborative node is connected to the sensor array and has a built-in data preprocessing unit, an outlier removal module, a collaborative decision-making module, and a device self-sustaining unit. The device self-sustaining unit triggers backflushing cleaning based on the pressure data from the micro pressure sensor. The dual-mode anti-interference collaborative transmission network adopts the LoRa-WiFi dual-mode wireless self-organizing network transmission module, which is suitable for inter-node collaborative communication, long-distance data transmission, and short-distance terminal interaction; a new dual-path cross-checking algorithm is added, which automatically switches to the backup link and corrects data deviation when strong winds cause signal interference. The working condition-physical field coupling acquisition unit integrates a wind speed sensor, an ambient temperature and humidity sensor, and a wind speed change rate calculation module to simultaneously acquire real-time wind force level and ambient temperature and humidity data of the construction environment. The intelligent data acquisition strategy optimization engine dynamically adjusts the data acquisition strategy based on wind speed level and physical field conditions.

3. The intelligent prediction and early warning system for roller-compacted concrete performance adapted to strong wind environments according to claim 2, characterized in that, The data preprocessing unit in the edge intelligent collaborative node is used for data normalization, noise reduction, and missing value completion; The outlier removal module sets a dynamic threshold based on the normal fluctuation range of the physical field under strong wind conditions to remove outlier data. The collaborative decision-making module is based on a distributed federated learning framework to build an edge node collaborative network. Each node only encrypts and shares data features and model parameters without leaking the original data. Through a preset voting mechanism, it makes collaborative judgments and dynamically optimizes the acquisition frequency and triggering timing of sensors across the entire area by combining on-site wind speed and concrete physical field conditions.

4. The intelligent prediction and early warning system for roller-compacted concrete performance adapted to strong wind environments according to claim 1, characterized in that, The wind speed gradient coupling unit in the strong wind coupled special spatiotemporal prediction model quantifies the nonlinear influence of strong wind on the physical field at different layers through a dynamic weight matrix, and establishes a mathematical relationship between wind speed and moisture evaporation, temperature and humidity conduction. The gradient mutation adaptive submodule sets up an LSTM decoder and a sliding window to monitor the first derivative of the physical field parameters. When the rate of change of the first derivative of the physical field parameters reaches a preset threshold, it activates local feature reinforcement learning. The layer-differentiated attention mechanism configures multiple attention heads at the surface, middle and bottom layers to capture the evolutionary specificity of each layer.

5. The intelligent prediction and early warning system for roller-compacted concrete performance adapted to strong wind environments according to claim 1, characterized in that, The causal dynamic early warning application layer includes: The multidimensional evolution trend display module is used to display real-time data of the physical field, future predicted evolution curves and subsequent performance prediction trends, and is configured with a strong wind impact analysis panel to quantify the contribution ratio of wind speed to each physical field index. The causal traceability multi-level early warning module triggers corresponding early warning signals based on the risk level and generates an early warning report that includes the risk level, evolution trend, root cause, scope of impact, and targeted control recommendations. The root cause is presented in the form of the contribution ratio of each factor. The adaptive control decision engine automatically matches the optimal control scheme library based on causal inference results, evaluates the effectiveness of measures through real-time data feedback, and dynamically iterates and optimizes control parameters. The system's self-optimization module is used to continuously collect real-time data and control effect data, periodically update the causal graph coefficients and dynamic threshold curves, and automatically generate construction parameter optimization suggestions. The causal graph coefficients are the core quantitative indicators that quantify the strength of causal relationships and contribution ratios among factors at each level in the "strong wind-construction parameters-physical field-subsequent performance" causal graph constructed by Bayesian network. The full lifecycle data management module stores monitoring data, forecast data, early warning data, operating condition data, and the implementation effects of control measures in a database, and supports querying and exporting by time range, level, and data type.

6. The intelligent prediction and early warning system for roller-compacted concrete performance adapted to strong wind environments according to claim 5, characterized in that, The automatic matching of the optimal control scheme library based on causal inference results specifically includes: When the root cause of the risk is excessive wind speed, a control plan should be implemented, including adding windproof barriers, surface spraying for moisturizing, and shortening the paving exposure time. When the root cause of the risk is the accumulation of hydration heat, then we will push for adjustments to the water cooling flow rate, surface ventilation, and optimized mix ratio control measures. When the root cause of the risk is insufficient maintenance, a control plan should be implemented to increase the frequency of watering the surface of the roller-compacted concrete and cover it with heat-insulating and moisture-retaining cotton blankets.

7. A method for intelligent prediction and early warning of roller-compacted concrete performance adapted to strong wind environments, applied to the system described in any one of claims 1 to 6, characterized in that, Includes the following steps: S1. Pre-construction preparation: During the construction preparation phase, the sensor array, edge nodes, transmission modules, and operating condition acquisition units are installed, debugged, and their parameters are initialized. S2. Dynamic acquisition and preprocessing: After the pouring is completed, the system is started. The edge intelligent collaborative monitoring layer dynamically adjusts the acquisition strategy according to the wind speed and physical field state to complete data preprocessing and anti-interference transmission. S3. Coupled Prediction and Causal Analysis: The strong wind-specific algorithm processing layer receives data, constructs a five-dimensional time series dataset, predicts the evolution of the physical field through a strong wind-coupled dedicated spatiotemporal prediction model, and evaluates the later performance and risk root causes based on the physical field-performance causal mapping module. S4. Dynamic Early Warning and Control: The causal dynamic early warning application layer triggers corresponding level early warnings based on prediction and evaluation results, pushes root cause analysis and control solutions, and receives execution feedback. S5. Adaptive Iterative Optimization: The system continuously collects real-time data and control effect data, iteratively updates model parameters, dynamic thresholds, and control schemes, forming a closed-loop optimization throughout the entire process.

8. The intelligent prediction and early warning method for roller-compacted concrete performance adapted to strong wind environments according to claim 7, characterized in that, The construction of the five-dimensional time series dataset in step S3 specifically includes: For the data received by the strong wind-specific algorithm processing layer, Z-score normalization is used to eliminate the influence of dimensions, and a five-dimensional time-series dataset containing wind speed, wind speed change rate, layer, time, and physical field is constructed. By dynamically assigning weights, the coupling effect of strong winds is highlighted, and the effectiveness of the dataset is optimized. The five-dimensional time series dataset is input into a strong wind coupled dedicated spatiotemporal prediction model to predict the evolution of the physical field.

Images